Power BI Authoring MCP Server
A local Microsoft MCP server for Power BI semantic models: an agent creates tables, DAX measures and relationships in Desktop, Fabric and PBIP files
- 1,240 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent creates, changes and deletes semantic model objects with your permissions, including in a Fabric workspace
- Model data and metadata reach the client and may go on to the LLM provider, and access tokens are sent to allowed hosts
Install
- Works in
Manual install
{"powerbi-authoring-local": {"type": "stdio", "command": "npx", "args": ["-y", "@microsoft/powerbi-modeling-mcp@latest", "--start"]}}For other MCP clients. Requires Node.js. Read the EULA and accept it before the other tools are used.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent creates, changes and deletes semantic model objects with your permissions, including in a Fabric workspace
- Model data and metadata reach the client and may go on to the LLM provider, and access tokens are sent to allowed hosts
This is third-party code. Review the repository files before installing.
What it does
The server gives an AI agent 21 tools for editing a Power BI semantic model. The model can live in Power BI Desktop, in a Fabric workspace or in PBIP project files with a TMDL definition. The agent creates and changes tables, columns, DAX measures, relationships, hierarchies, calculation groups, row-level security roles, perspectives and translations. Separate tools handle transactions, partitions and Power Query parameters. The server also runs and validates DAX queries. Built-in prompts CreateDAXQuery, RunDAXQueryWithMetrics and AnalyzeDAXQuery cover that, and ConnectToPowerBIDesktop, ConnectToFabric and ConnectToPBIP prompts connect to a model. Bulk operations such as renaming hundreds of objects run in batches with transactions. The server runs locally and connects to the model over XMLA with your permissions. By default it works in read-write mode and asks for confirmation before the first edit in each database. The --readonly flag blocks any changes. The server edits the model only and does not touch report pages.
Who it is for. For BI developers and analysts who build semantic models in Power BI and want Copilot or another agent to handle renaming, documentation and DAX measure refactoring.
Good fit when
- A model has dozens of tables with mixed naming styles, and the agent should analyze the conventions and bulk rename objects
- You need descriptions for all measures, columns and tables, with the DAX logic explained in plain terms
- The model needs a French or other translation: the culture and object-translation tools do it in one pass
- Several similar measures such as 6 and 12 month rolling averages should become a calculation group with new 24M and 3M variants
- You want to benchmark one DAX query across two model versions and review the execution metrics
Not a fit when
- The model is in a Fabric workspace: Microsoft itself recommends the remote hosted Power BI Authoring MCP server there, with no local install
- You need to change report pages or the model's diagram layout: the server performs modeling operations only
- You lack Write permission on the semantic model: edits will fail without it
- There is no model backup and blind edits by an agent are not acceptable: the author advises making a backup first
Example request
Connect to 'Sales.pbix' in Power BI Desktop, analyze the naming conventions and suggest renames for the Sales table.MCP
- Transport
- stdio
- Authentication
- OAuth
Environment variables
AZURE_CLIENT_ID- Microsoft Entra application ID, required with --authmode=serviceprincipal.
AZURE_TENANT_ID- Microsoft Entra tenant ID.
AZURE_CLIENT_SECRETsecret- Service principal secret when signing in with a secret.
PBI_MODELING_MCP_ACCEPT_EULA- Set to true to accept the EULA for unattended execution.
PBI_MODELING_MCP_ALLOWED_CONNECTION_HOSTS- Trusted Analysis Services or XMLA hosts that the server may send access tokens to.
Limitations
- Write permission on the semantic model is required. The server connects through the XMLA endpoint.
- Before using the npm package you must accept the EULA: until you do, the server blocks all other tools. For unattended runs there is the --accepteula flag and the PBI_MODELING_MCP_ACCEPT_EULA variable.
- Model data and metadata go to the client and may reach the LLM provider.
- The software may collect usage data and send it to Microsoft.
- No tenant setting blocks the server: access can be stopped only by disabling the XMLA endpoint.
How to disable. In VS Code disable or remove the Power BI Authoring MCP extension. For a manual install delete the powerbi-authoring-local entry from the client's MCP config. To make the server read-only instead of removing it, add the --readonly argument.
FAQ
How do I connect the Power BI MCP server to VS Code and GitHub Copilot?
Install Visual Studio Code, the GitHub Copilot Chat extension and the Power BI Authoring MCP extension. Then confirm in Copilot chat that powerbi-modeling-mcp appears in the tool list. If it does not, check that the MCP servers in Copilot setting is enabled; for enterprise accounts an administrator turns it on.
How do I run the Power BI Authoring MCP server with npx?
Add a stdio server to the client config with the command npx and the arguments -y, @microsoft/powerbi-modeling-mcp@latest, --start. Until the EULA is accepted all other tools are blocked: the agent may call accept_eula only after your explicit permission.
Can the agent break my Power BI model?
It can, if you have write permission. By default the server asks for confirmation before the first edit in each database, and the --readonly flag blocks any changes. The author advises making a model backup before you start.
Does the server work with Fabric models and PBIP files?
Yes. You can connect to a file in Power BI Desktop, to a semantic model in a Fabric workspace and to the definition folder of a PBIP project. For Fabric models Microsoft recommends the remote server, which needs no installation.
Which tools does the Power BI Authoring MCP Server include?
Twenty-one tools, from connection_operations and database_operations to measure_operations, relationship_operations, dax_query_operations, security_role_operations and object_translation_operations. The agent can explain them itself when you ask "Tell me with some examples what I can do with the Power BI Authoring MCP server".
Discussion
No questions or reviews yet.
Related
MCP serversMCP server and CLI to drive real Microsoft Excel: Power Query, DAX, VBA, PivotTables and charts
A CLI and 13 skills for Claude Code: the Power BI data model, DAX measures, row-level security and PBIR report pages straight from chat
A plugin marketplace for Claude Code and GitHub Copilot on Power BI and Microsoft Fabric: semantic models, DAX, TMDL, reports
dbt MCP
dbt-mcp
The official MCP server from dbt Labs: the Semantic Layer, project metadata, dbt commands and SQL against the warehouse from an agent
The official MathWorks MCP server: the agent starts and quits MATLAB, writes and runs code, and checks it for errors and style
Official YDB MCP server
YDB MCP
YDB's own official MCP server from the database's developers: parameterized SQL queries, execution plans, directory browsing, and a base for custom servers
README in short
The README opens with a note that models in Fabric should use Microsoft's remote server. It then covers capabilities: natural-language edits, bulk operations, best-practice checks, TMDL support and DAX queries. The install section describes the VS Code extension, launch through npx and manual unpacking of the VSIX package. Tools, prompts, command-line options and environment variables follow. The end collects warnings about backups, data privacy, permissions and telemetry.